-
Notifications
You must be signed in to change notification settings - Fork 442
All issues
Issue creation is restricted in this repository
- #1603 · SolitaryThinker opened
on Jul 15, 2026 1
Issues
is:issue state:open
is:issue state:open
Search results
Support FSDP inference with precomputed quantized weights
scope: attentionAttention backends (VSA, STA, Flash, etc.)Attention backends (VSA, STA, Flash, etc.)scope: distributedSP, FSDP, USP, multi-nodeSP, FSDP, USP, multi-nodescope: inferenceInference pipeline, serving, CLIInference pipeline, serving, CLIscope: trainingTraining pipeline, methods, configsTraining pipeline, methods, configsStatus: Open.Support LoRA lifecycle operations for MXFP8 and NVFP4 inference
scope: inferenceInference pipeline, serving, CLIInference pipeline, serving, CLIscope: trainingTraining pipeline, methods, configsTraining pipeline, methods, configsStatus: Open.#1821 In hao-ai-lab/FastVideo;[Bug] Cosmos AdaLayerNorm autocast hardcodes device_type="cuda", breaking on non-CUDA accelerators
installationInstallation and setup issuesInstallation and setup issuesplatformPlatform-specific (Windows/macOS)Platform-specific (Windows/macOS)scope: modelModel architecture (DiTs, encoders, VAEs)Model architecture (DiTs, encoders, VAEs)Status: Open.#1816 In hao-ai-lab/FastVideo;[Bug] VBench SceneMetric loads Qwen2.5-Omni on CPU on any non-CUDA accelerator (device_map hardcodes cuda)
installationInstallation and setup issuesInstallation and setup issuesplatformPlatform-specific (Windows/macOS)Platform-specific (Windows/macOS)scope: inferenceInference pipeline, serving, CLIInference pipeline, serving, CLIStatus: Open.#1815 In hao-ai-lab/FastVideo;Quickstart example OOMs on 16GB-RAM free-tier environments (Kaggle, Colab)
installationInstallation and setup issuesInstallation and setup issuesperformancePerformance and memory issuesPerformance and memory issuesscope: attentionAttention backends (VSA, STA, Flash, etc.)Attention backends (VSA, STA, Flash, etc.)scope: docsDocumentationDocumentationscope: modelModel architecture (DiTs, encoders, VAEs)Model architecture (DiTs, encoders, VAEs)Status: Open.#1755 In hao-ai-lab/FastVideo;[bug] Qwen3-VL vision tower diverges from transformers 5.15.0: visual-token embeddings mismatch (image/video conditioning parity broken on main)
platformPlatform-specific (Windows/macOS)Platform-specific (Windows/macOS)scope: attentionAttention backends (VSA, STA, Flash, etc.)Attention backends (VSA, STA, Flash, etc.)scope: inferenceInference pipeline, serving, CLIInference pipeline, serving, CLIStatus: Open.#1733 In hao-ai-lab/FastVideo;[kernel] [Feature] NVFP4 QAT: fix broken measurement plumbing and build MFU foundations (pre-work for DGX Spark QAT kernel optimization)
installationInstallation and setup issuesInstallation and setup issuesplatformPlatform-specific (Windows/macOS)Platform-specific (Windows/macOS)scope: docsDocumentationDocumentationscope: kernelCUDA kernels, fastvideo-kernelCUDA kernels, fastvideo-kernelscope: trainingTraining pipeline, methods, configsTraining pipeline, methods, configsStatus: Open.#1722 In hao-ai-lab/FastVideo;- Status: Open.#1713 In hao-ai-lab/FastVideo;
[Bug] MiniMax H3 text encoder needs ~97 GB resident to load a 63 GB bf16 checkpoint(FAS-494)
installationInstallation and setup issuesInstallation and setup issuesperformancePerformance and memory issuesPerformance and memory issuesplatformPlatform-specific (Windows/macOS)Platform-specific (Windows/macOS)scope: attentionAttention backends (VSA, STA, Flash, etc.)Attention backends (VSA, STA, Flash, etc.)scope: docsDocumentationDocumentationscope: inferenceInference pipeline, serving, CLIInference pipeline, serving, CLIscope: modelModel architecture (DiTs, encoders, VAEs)Model architecture (DiTs, encoders, VAEs)Status: Open.#1709 In hao-ai-lab/FastVideo;[Bug] HunyuanVideo 1.5 I2V cannot run: image_embeds is never populated, stage verification fails
installationInstallation and setup issuesInstallation and setup issuesperformancePerformance and memory issuesPerformance and memory issuesplatformPlatform-specific (Windows/macOS)Platform-specific (Windows/macOS)scope: attentionAttention backends (VSA, STA, Flash, etc.)Attention backends (VSA, STA, Flash, etc.)scope: distributedSP, FSDP, USP, multi-nodeSP, FSDP, USP, multi-nodescope: docsDocumentationDocumentationscope: inferenceInference pipeline, serving, CLIInference pipeline, serving, CLIscope: modelModel architecture (DiTs, encoders, VAEs)Model architecture (DiTs, encoders, VAEs)scope: trainingTraining pipeline, methods, configsTraining pipeline, methods, configsStatus: Open.#1692 In hao-ai-lab/FastVideo;[RFC]: Command "dreamverse-server --host 0.0.0.0 --port 8009": GPU 0 warmup failed
installationInstallation and setup issuesInstallation and setup issuesscope: attentionAttention backends (VSA, STA, Flash, etc.)Attention backends (VSA, STA, Flash, etc.)scope: inferenceInference pipeline, serving, CLIInference pipeline, serving, CLIscope: trainingTraining pipeline, methods, configsTraining pipeline, methods, configsStatus: Open.#1685 In hao-ai-lab/FastVideo;[ci] [dashboard]: add explicit benchmark cohort comparison mode
platformPlatform-specific (Windows/macOS)Platform-specific (Windows/macOS)scope: attentionAttention backends (VSA, STA, Flash, etc.)Attention backends (VSA, STA, Flash, etc.)Status: Open.#1636 In hao-ai-lab/FastVideo;